异号权重模板CNN的似细化特性及其应用

Thinning-Like Property of CNN with Opposite-Sign Templates and lts Applications

本文发现并以定理的形式证明了具有异号权重模板的细胞神经网络（ＣＮＮ）系统在非均匀增益分段线性输出函数下的似细化稳态特性．利用这似细化特性设计了一多层带有局部逻辑单元的ＣＮＮ网络实现图像细化．还设计了另一并行ＣＮＮ网络实现图像边缘检测．文中给出了模拟实验结果．

In this paper,a thinning-like (TL) property of cellular neural networks(CNN)with opposite-sign templates and a non-unity gain output function is discovered and proved. An image thinning algorithm is realized by a multi-layer CNN,which operates with the TL property,with four local logic units (LLU). An image edge detection is achieved by another parallel CNN with a LLU. Some simulation results are given.